CLIP Text Encode (Prompt) node in ComfyUI

CLIP Text Encode turns a text prompt into CONDITIONING using the model text encoder. One node holds the positive prompt, a second one the negative prompt.

Node pack: ComfyUI core

Category: Conditioning and prompts

What CLIP Text Encode (Prompt) does

The sampler cannot read words. CLIP Text Encode runs the prompt through the CLIP (or T5) text encoder and produces the conditioning tensors that steer the denoising. A standard workflow has two of them: the positive prompt wired to the KSampler positive input and the negative prompt wired to negative.

The text box supports ComfyUI prompt syntax: (word:1.3) raises attention on a word, (word:0.7) lowers it, plain parentheses multiply by 1.1, and embeddings are referenced as embedding:name. Prompts longer than 75 tokens are split into chunks automatically, and BREAK starts a new chunk by hand.

Inputs

Name	Type	What it is
clip	CLIP	The text encoder from Load Checkpoint, Load LoRA or DualCLIPLoader.
text	STRING	The prompt. Multi-line; can be converted to an input to feed it from another node.

Outputs

Name	Type	What it is
CONDITIONING	CONDITIONING	The encoded prompt for KSampler, ControlNet Apply or conditioning combine nodes.

Skip the setup: ComfyUI workflows preinstalled on the cloud

BitVector runs ready-made ComfyUI workflows on its own GPUs. No install, no missing nodes, no red boxes. Open it on your phone or laptop and generate in a minute.

Try BitVector

How to use CLIP Text Encode (Prompt)

Add two CLIP Text Encode nodes and connect the CLIP output of your loader (after any LoRA) to both.
Write the positive prompt in one and unwanted things in the other.
Connect the first CONDITIONING to KSampler positive and the second to negative.
Use (term:1.2) weights sparingly; 1.1 to 1.4 is the useful range.

Settings and tips

Put the most important words first; CLIP pays more attention to the start of each 75-token chunk.
Textual inversion embeddings: write embedding:filename (without extension) in either box; negative embeddings like easynegative go in the negative box.
For Flux and SD3 write natural sentences; weights and tag soups work worse with T5.
Right-click the node and choose "Convert text to input" to drive the prompt from a text node or a wildcard node.
An empty negative prompt is fine; ComfyUI encodes an empty string.

Troubleshooting CLIP Text Encode (Prompt)

Token indices sequence length is longer than the specified maximum (77)

Why it happens

The prompt exceeds 75 tokens. ComfyUI splits it into chunks, the message is only a warning from the tokenizer.

How to fix it

Nothing to fix for correctness. Shorten the prompt or use BREAK to control where the chunks split if later words get ignored.

The prompt seems ignored and the image barely changes

Why it happens

The node is wired to a CLIP that does not match the model (for example the checkpoint CLIP while the model goes through a LoRA), or cfg is 1 on an SD model so guidance has no effect.

How to fix it

Wire the CLIP from the last loader in the chain and keep cfg between 4 and 8 for SD 1.5 and SDXL. For Flux, use Flux Guidance instead of cfg.

Error: embedding not found / "embedding:xyz" printed in the output

Why it happens

The embedding file is missing from models/embeddings or the name has a typo or extension.

How to fix it

Copy the .pt or .safetensors file to models/embeddings and reference it by file name without extension, matching case on Linux.

Weighted words like (red:1.5) produce artifacts

Why it happens

Weights above about 1.5 push the embedding outside the range the model was trained on.

How to fix it

Keep weights under 1.4 and repeat the concept in words instead of raising the number.

Run this workflow from your phone

Every workflow on this page is preinstalled on BitVector with the models and custom nodes already in place. Pick one, type a prompt, done. Zero setup, nothing to download.

Try BitVector

Questions about CLIP Text Encode (Prompt)

Do I need a negative prompt?

For SD 1.5 and SDXL it helps a lot (blurry, lowres, bad hands). Flux at cfg 1 ignores it entirely.

Why are there "CLIP Text Encode (Prompt)" and "CLIPTextEncodeSDXL" nodes?

The SDXL variant lets you set separate texts for the two SDXL encoders and target sizes. The plain node works fine for SDXL in practice.

What does BREAK do?

It ends the current 75-token chunk so the words after it start a fresh chunk with full attention.

Related nodes

Load Checkpoint

ComfyUI core

Load Checkpoint (CheckpointLoaderSimple) opens a .safetensors or .ckpt model file and hands out the three parts every workflow needs: the diffusion MODEL, the CLIP text encoder and the VAE.

KSampler

ComfyUI core

KSampler runs the denoising loop: it takes the model, prompts and a latent and produces the finished latent image, controlled by seed, steps, cfg, sampler, scheduler and denoise.

Conditioning (Combine) and (Concat)

ComfyUI core

Conditioning (Combine) merges two conditionings so the sampler is guided by both at once; Conditioning (Concat) joins the token sequences into one longer prompt. Related nodes set area, mask and timestep ranges.

CLIP Set Last Layer

ComfyUI core

CLIP Set Last Layer is the ComfyUI equivalent of "clip skip": it stops the text encoder a few layers early (-2 is the common setting for anime and Pony models).

Flux Guidance

ComfyUI core

Flux Guidance (FluxGuidance) writes the guidance value into the conditioning that Flux Dev was distilled to expect, replacing cfg; 3.5 is the default and 2 to 5 the useful range.

Tired of fixing nodes? Let the cloud do it

BitVector keeps hundreds of ComfyUI workflows installed, updated and tested on fast cloud GPUs. No Python, no CUDA errors, no VRAM limits. Works from any browser.

Try BitVector